Matrix mechanics for actual atoms and molecules

Alexei M. Frolov∗
Department of Applied Mathematics
University of Western Ontario, London, Ontario N6H 5B7, Canada
(Dated: August 11, 2017)
arXiv:1708.03129v1 [quant-ph] 10 Aug 2017

Abstract
Matrix mechanics is developed to describe the bound state spectra in few- and many-electron
atoms, ions and molecules. Our method is based on the matrix factorization of many-electron (or
many-particle) Coulomb Hamiltonians which are written in hyperspherical coordinates.
PACS number(s): 31.30.Gs, 31.15.vj and 32.15.Fn

∗
E–mail address: [email protected]

1
I. INTRODUCTION

In this communication we develop the matrix mechanics of the actual, i.e. few- and
many-electron, atoms, ions and molecular systems. This approach is, in fact, a very pow-
erful method for analysis of various few- and many-electron Coulomb systems which can
succsessfully be applied to describe the bound state spectra in different atoms, ions and
even molecules. Our approach is a new step in the development of Matrix Mechanics [1]
which was originally created by Heisenberg, Born and Jordan as the first version of Quantum
Mechanics [2], [3]. Briefly, we want to show how the old version of matrix mechanics can be
modified to the new level and can be used as an effective tool for solving numerous problems
in modern atomic physics.
Our main goal in this study is to show that the Coulomb Hamiltonian of an arbitrary
atom which contain Ne −bound electrons is always factorized, i.e. it is represented in the
form of a product of the two differential operators of the first order. This fundamental fact is
directly related to the internal structure of the Coulomb Hamiltonians (the so-called ladder
structure) and substantially simplifies analysis of the bound state spectra in few- and many-
electron atoms and ions and can be used to perform more accurate numerical computations
of the bound states. In particular, by using the method of matrix factorization we can
determine the energies and wave functions of an arbitrary bound state in many-electron
atoms and ions, including excited and highly excited bound states. The same procedure can
also be used for molecules and for other many-particle Coulomb systems.
First, let us consider the one-electron hydrogen atom and/or hydrogen-like ions, i.e.
atomic systems which contain one bound electron and one positively charged nucleus. To
simplify our analysis in this study we shall assume that all atomic nuclei mentioned below
are infinitely heavy. Furthermore, everywhere below we shall apply the atomic system of
h
units. In these units h̄ = 1, | e |= 1 and me = 1, where h̄ = 2π
is the reduced Planck
constant, me is the electron mass and e is the electric charge of electron (a negative value).
In atomic units the Hamiltonian of one-electron atoms/ions is written in the form

h̄2 h ∂ 2 2 ∂ L2 i Qe2 1 h ∂2 2 ∂ L2 i Q
H=− + − − = − + − − (1)
2me ∂r 2 r ∂r r2 r 2 ∂r 2 r ∂r r2 r

where Qe = Q is the electric charge of the atomic nucleus and L is the operator of the angular
moment of the atom which coincides with the total angular momentum of the bound atomic

2
electrons. To determine the bound states in the hydrogen atom and hydrogen-like ions we
need to solve the corresponding Schrödinger equation HΨ = EΨ, where the operator H
is the Hamiltonian, Eq.(1), Ψ is the unknown wave function and E is the eigenvalue of H
which is the total energy of the bound state, i.e. E < 0. As is well known (see, e.g., [4],
[5]) the total wave function of an arbitrary bound state of the hydrogen atom is represented
as a product of the radial part of the total wave function ψnℓ (r) and the corresponding
spherical harmonic(s) Yℓm (θ, φ), i.e. Ψnℓm (r, θ, φ) = ψnℓ (r)Yℓm (θ, φ), where Yℓm (θ, φ) are
the eigenfunctions of the L2 operator, i.e. L2 Yℓm (θ, φ) = ~ℓ2 Yℓm (θ, φ) = ℓ(ℓ + 1)Yℓm (θ, φ).
Here and everywhere below the notations θ and φ stand for the spherical coordinates of
the bound electron, while r is the electron-proton distance which coincides with the radial
spherical coordinate. The integer numbers n, ℓ and m are called the principal quantum
number, angular quantum number and magenetic quantum number, respectively. For one-
electron atomic systems all these quantum numbers are the ‘good’ (or conserving) quantum
numbers. Note also that the following inequalies are always obeyed for these quantum
numbers: ℓ ≤ n − 1 and | m |≤ ℓ.
In the basis of spherical harmonics Yℓm (θ, φ) the Hamiltonian, Eq.(1), takes the form

1 h ∂2 2 ∂ i ℓ(ℓ + 1) Q n 1 h ∂2 2 ∂ i ℓ(ℓ + 1) Q o
H(r) = − + + − = δℓ,ℓ δm,m − + + − (2)
2 ∂r 2 r ∂r 2r 2 r 1 1
2 ∂r 2 r ∂r 2r 2 r

where ℓ is the angular moment of the bound electron (ℓ ≥ 0) which coincides (for one-
electron atoms/ions) with the angular momentum of the whole atom L. Note that the
Hamiltonian, Eq.(2) is a differential operator of the second order upon the radial variable
r. On the other hand, the Hamiltonian H(r) is a diagonal matrix in terms of the ℓ and m
(or | m |) indeces, or conserving quantum numbers. The explicit solution of the Schrödinger
equation Hψ = Eψ for the bound states of the hydrogen atom and hydrogen-like ions leads
to the following formula (Borh’s formula) for the energy spectrum

me Q2 e4 Q2 Q2
En = − = − = − (3)
2h̄2 n2 2n2 2(nr + ℓ + 1)2

where nr is the radial quantum number which is a non-negative integer and varies between
0 and n − ℓ − 1 and n is the principal quantum number. The numerical value of nr coincides
with the number of zeros in the radial part of the wave function ψnℓ (r). Furthermore, the
radial part of the total wave function ψnℓ (r) equals to the product of some positive power
of r, Laguerre polynomial of r and a radial exponent. In numerous textbooks this results

3
is derived by using a special form of the radial wave function ψnℓ (r) (see, e.g., [5]). Then
the original differential equation is reduced to the corresponding differential equation for the
hypergeometric function 2 F1 (a, b; c; r) which must have a finite number of terms, or, in other
words, to be a polynomial. This is the standard procedure which have been described in
many textbooks. However, there is another procedure which can be applied to determine the
bound state spectrum, i.e. the total energies and wave functions, of the hydrogen atom and
hydrogen-like ions. This procedure is more elegant, physically transparent and based on the
internal structure of the Coulomb Hamiltonian (see, e.g., [1]). We describe this procedure
in the next Section.

II. FACTORIZATION METHOD FOR ONE-ELECTRON ATOM/ION

This Section is intended merely to summarize the central facts about the factorization
method that are needed in Sections IV - V below. Another aim of this Section is to fix the
notation. Now, consider the matrix of the Hamiltonian H, Eq.(1), in the basis of spherical
harmonics, i.e. the matrix hYℓm(θ, φ) | H | Yℓ1 m1 (θ, φ)i = δℓ,ℓ1 δm,m1 Ĥℓ,m (r) = H(r) which is
a diagonal matrix in the ℓ and m indeces. On the other hand, each matrix element of this
matrix is a differential operator of the radial variable r, i.e. Ĥℓ,m (r). Since the both ℓ and m
quantum numbers are the conserving (or ‘good’) quantum numbers, then we can replace the
corresponding matrix notation Ĥℓ,m (r) by a simple operator notation, i.e., Ĥℓ,m (r) = H(r)
(see, Eq.(2)). Our goal in this Section is to find all eigenvalues of this radial operator
H(r), Eq.(2). For these purposes we shall apply the factorization method developed for
the differential operators of the second order. This method was well described in a number
of books and textbooks (see, e.g., [1], [6] and references therein). Below, we assume that
the reader is acquainted with the factorization method and its applications to one-electron
atomic systems (see, e.g., [1] and references therein).
The method of matrix factorization (see, e.g., [1]) is based on the existence of a set of
the first-order differential operators Θn (r) (where n = 1, 2, . . .) and their adjoint operators
Θ∗n (r). The Θn (r) operators are written in the form

1 h ∂ 1  βn i
Θn (r) = √ − + + + αn (4)
2 ∂r r r

4
In respect to this definition the adjoint operators are
1 h ∂ 1  βn i
Θ∗n (r) = √ + + + αn (5)
2 ∂r r r
The real parameters βn and αn in operators defined by Eqs.(4) - (5) must be choosen to
obey the two fundamental conditions of the factorization method. First, the Hamiltonian
H(r), Eq.(2), must be represented in the form

H = Θ∗1 (r)Θ1 (r) + a1 (6)

where H is the Coulomb Hamiltonian, Eq.(2), of the one-electron hydrogen atom. Second,
there is an infinite, in principle, chain of relations between the Θn (r), Θ∗n (r), Θ∗n+1(r) and
Θn+1 (r) operators:

Θn (r)Θ∗n (r) + an = Hn+1 = Θ∗n+1 (r)Θn+1 (r) + an+1 (7)

where Hn+1 is the n−excited Hamiltonian (or n-times excited Hamiltonian, where n ≥ 1)
of the original problem. In this notation we have to assume that H1 = H. The equations,
Eqs.(6) - (7), and their role in the factorization method are discussed in detail in [1]. In this
study we do not want to repeat that description. Instead, we note that from Eq.(6) and
explicit formulas, Eqs.(4) and (5), written for n = 1, one finds three following equations for
the β1 , α1 and a1 parameters
1
β1 (β1 − 1) = ℓ(ℓ + 1) , α1 β1 + β1 α1 = 2β1 α1 = 2Q , a1 = − α12 (8)
2
From the first equation we obtain β1 = ℓ + 1. Another solution which corresponds to the
β1 = −ℓ value cannot be accepted, since it produces the wave function which is singular at
the radial origin, i.e. at r = 0. Such solutions have no physical sense for the Coulomb two-
Q
body problem. By using the relation β1 = ℓ + 1 we deternine the parameter α1 : α1 = ℓ+1
.
Q 2
Then from Eq.(8) one finds that a1 = − 2(ℓ+1)2 . This expression for the parameter a1 exactly

coincides with total energy of the lowest bound state in a series of bound states with the
angular momemtum ℓ.
Analogously, by substituting the experssions, Eqs.(4) - (5), into the formula for the Hn+1
Hamiltonian, Eq.(7), we obtain the following equations for the αn , βn , αn+1, βn+1 and an+1
values
1 2
βn+1 (βn+1 − 1) = βn (βn + 1) , 2αn+1 βn+1 = 2Q = 2αn βn , an+1 = − αn+1 (9)
2

5
 Q
From these equations one finds that βn+1 = βn + 1 = . . . = β1 + n = n + ℓ + 1, αn+1 = n+ℓ+1
Q 2
and an+1 = − 21 αn+1
2
= − 2(n+ℓ+1)2 . This value of an+1 exactly coincides with the total energy

of the n−excited bound state (En+1 ) in the series of bound states with the given value of ℓ.
In other words, by using this simple method one can reproduce the bound state spectra for
the series of bound states with arbitrary ℓ (angular momentum). It follows from here that
the factorization method also produces the whole bound state spectrum of the hydrogen
atom which contains the bound states with different values of angular momentum ℓ (ℓ ≥ 0).
Now, consider the energy functional E(Ψ)(= E1 (Ψ)) (see, e.g., [7]), where Ψ = Ψ(r) is
the trial function, and the Hamiltonian H is represented in the form of Eq.(6)
hΨ | H | Ψi hΨ | Θ∗1 (r)Θ1 (r) | Ψi hΘ1 (r)Ψ | Θ1 (r)Ψi
E(Ψ) = = + a1 = + a1 (10)
hΨ | Ψi hΨ | Ψi hΨ | Ψi
where a1 is some number negative number which is uniformly defined by Ψ. Since the
first term in the right-hand side of this equation is always positive, then it follows from
Eq.(10) that minΨ E(Ψ) = a1 and such a minimum is reached on the function Ψ which is
defined by the equation Θ1 (r)Ψ(r) = 0. Thus, we have found the equation which allows one
to obtain the ground state wave function Ψ1 (r) of an arbitrary one-electron atom and/or
ion. At the next step we consider the subspace of functions Φ which are represented in the
form Φ(r) = Θ∗1 (r)Ψ(r), where the function Ψ is an arbitrary radial function defined in the
L2 (0 ≤ r < ∞) space. It is clear any of these functions is orthogonal to the ground state
wave function Ψ1 (r), since hΘ∗1 (r)Ψ(r) | Ψ1 (r)i = hΨ(r) | Θ1 (r)Ψ1 (r)i = 0. This means
that we are dealing with the subspace of the trial functions which are represented in the
form Φ(r) = Θ∗1 (r)Ψ(r) and all these functions Φ(r) are orthogonal to the ground state wave
funcition Ψ1 (r).
For the Φ(r) functions we can investigate the following energy functional
hΦ | Φi hΘ∗1 (r)Ψ | Θ∗1 (r)Ψi
F (Φ, Ψ) = F (Ψ) = = + a1 (11)
hΨ | Ψi hΨ | Ψi
By using the equality, Eq.(7), one can reduce this functional to the form
hΨ | Θ∗2 (r)Θ2(r) | Ψi hΘ2 (r)Ψ | Θ2 (r) | Ψi
F (Ψ) = + a2 = + a2 = E2 (Ψ) (12)
hΨ | Ψi hΨ | Ψi
where E2 is the variational energy of the first excited state of the hydrogen atom and a2
is a real negative number. It is clear that the minimum of the functional F (Ψ) = E2 (Ψ),
Eq.(12), equals to the a2 value which coincides with the total energy of the first excited

6
state. The corresponding eigenfunction is defined by the equations: Θ2 (r) | Ψi = 0 and
Θ1 (r) | Ψi =
6 0. Note that our trial functions used in Eq.(12) are already ‘correct’ trial
functions, since they do not have any non-zero component which is proportional to the
ground state wave function Ψ1 . Briefly, we can say that the minimum of the ‘excited’
energy functional equals a2 , Eq.(12), while the corresponding wave functions are obtained
from the equations Θ2 (r) | Ψi = 0.
Then we can repeat this procedure by considering the non-zero functions represented in
the form Θ∗2 (r)Φ(r) = Θ∗2 (r)Θ∗1(r)Ψ(r), where the function Ψ(r) is an arbitrary, in principle,
radial function defined in the L2 (0 ≤ r < ∞) space. Then, with the help of Eqs.(7) and (9)
the whole process can be repeated again as many times as needed to determine all energies of
the bound states and their wave functions. The explicit form of the ground state wave func-
Q
tion for one-electron atom/ion with our values of β1 and α1 is Ψ1 (r) = Cr ℓ exp(− ℓ+1 r) where
C is the normalization constant. This function is the well known exact wave function of the
lowest (by the energy) state in the series of bound states with the given value of angular mo-
mentum ℓ (see, e.g., [6]). In general, the radial wave function Ψn of the n−excited state can
be determined from the equation Θn (r)Ψn (r) = 0. Such a wave function must be orthogonal
to the corresponding radial wave functions Ψn−1 (r), Ψn−2(r), . . . , Ψ1 (r) of all lower bound
states. This means that we can consider the radial functions Ψn (r), Ψn−1 (r), . . . , Ψ1 (r) as a
‘basis’ in the n−dimensional subspace in the L2 (0 ≤ r < ∞) space of the radial functions.
As is well known such a basis in n−dimensional space can be orthogonalized by using a
simple procedure which described in detail in many textbooks (see, e.g., [8], [9]). After
orthogonalization we obtain the system of unit-norm radial functions which exactly coincide
with the known radial functions of the hydrogenic systems (see, e.g., [5], [10]).

III. METHOD OF HYPERSPHERICAL HARMONICS

This and two following Sections are the central part of our study, since here we generalize
the factorization method to the new level in order to include applications to various few- and
many-electron atoms and ions, or, in other words, to many-particle Coulomb systems. Let Ne
be the total number of bound electrons in such an atomic system. The approach described in
the previous Section works only for one-electron atomic systems, i.e. for Ne = 1. For atomic
systems which contain two, three, and/or more bound electrons we need to develop the new

7
approach and introduce a convenient system of new notations. First, it is clear that the
total number of spatial variables in the case of many-electron atoms is substantially larger
than three and we need to use more variables to designate all electron’s spatial coordinates.
This problem is solved below by introducing the complete set of 3Ne electron hyperspherical
coordinates. There are also Ne electron spin coordinates which are combined in the total
electron spin S (or S(S + 1) value and its z−projection Sz (see below). Second, it is also
a priori clear that the complete sets of conserving quantum numbers (or sets of ‘good’
quantum numbers) are substantially different for one- and few-electron atomic systems. In
particular, the angular momentum of any single bound electron ℓi , where i = 1, 2, . . . , Ne , in
many-electron atoms is not conserved. However, the vector-sum of the angular momenta of
all bound electrons L = ~ℓ1 + ~ℓ2 +. . . ~ℓNe is conserved. Analogously, for a single atomic electron
the projection of its angular moment at z−axis, i.e. ℓzi (= mi ) value, is not conserved, while
the sum Lz = ℓ1,z + ℓ2,z + . . .+ ℓNe ,z = m1,z + m2,z + . . .+ mNe ,z = M is a conserving (or good)
quantum number which is often called the magnetic quantum number M. In general, for
an arbitrary bound state in many-electron, non-relativistic atomion one finds the following
set of conserving quantum numbers L, M and π, where π = (−1)ℓ1 +ℓ2 +...+ℓNe is the spatial
parity of the atomic wave function, or spatial parity of the bound state. In addition to
these three quantum numbers in any isolated atomic system with bound electrons one finds
the two additional quantum numbers which are always conserved: (1) the total electron
spin S (or S(S + 1)), and (2) the projection of the total electron spin S on the z−axis
which is designated below as Sz [5]. The set of these five integer and semi-integer numbers
h i
L, M, S, Sz , π uniformly defines one series of bound atomic states which is usually called
the atomic term (for more details, see, e.g., [10]).
In atomic physics the hyperspherical coordinates were introduced by Fock in 1954 [11]
when he investigated the bound state wave function of the ground 11 S−state in the two-
electron He atom. Later these coordinates were used in accurate computations of the dif-
ferent bound states of the He atom [12]. Since the middle of 1960’s the hyperspherical
coordinates have extensively been used in nuclear and hyper-nuclear few-body problems. It
was found that such coordinates are appropriate to describe various few-body systems which
are close to their dissociation threshold(s). In 1974 Knirk [13] re-introduced the new set of
hyperspherical coordinates in atomic and molecular physics. The choice of the hyperspher-
ical coordinates in atomic problems with Ne −bound electrons made by Knirk was different

8
from that used earlier by Fock [11]. We have found that the definition of the hyperspherical
coordinates proposed by Knirk (see Section II of his paper [13]) is more convenient and
appropriate for various atomic problems.
In this study, we shall use the same hyperspherical coordinates which exactly coincide
with such coordinates defined in [13]. In particular, the angular (or spherical) coordinates
of each electron are designated below as ωi = (θi , φi ), where i = 1, 2, . . . , Ne . The radial
variables of each electron ri are defined exactly as in Eq.(2.1) from [13] and hyper-radius r
coincides with the expression given in Eq.(2.3) from [13]. In other words, we can write for
the Cartesian coordinates of each electron

xi = ri sin θi cos φi , yi = ri sin θi sin φi , zi = ri cos θi , (13)

where i = 1, 2, . . . , Ne , while (xi , yi , zi ) are the Cartesian coordinates of the i−th electron
q
and ri = x2i + yi2 , +zi2 is the spherical radial cooordinate of this electron. It is clear that
~ℓ2 Yℓm (θj , φj ) = −∆i Yℓm (θj , φj ) = δij ℓ(ℓ+1)Yℓm(θi , φi ), where ~ℓ2 is the square of the ordinary
i i

momentum operator of the particle i.

qP
Ne 2
Now, we can define the atomic hyper-radius r = i=1 ri and (Ne − 1) hyperspherical
angles η2 , η3 , . . . , η3Ne −1 which are defined by the following relations

rNe = r cos ηNe , rNe −1 = r sin ηNe cos ηNe −1 , rNe −2 = r sin ηNe sin ηNe −1 cos ηNe −2 , . . . ,
r2 = r sin ηNe sin ηNe −1 . . . sin η3 cos η2 , r1 = r sin ηNe sin ηNe −1 . . . sin η3 sin η2 (14)

The set of (3Ne − 1) angular variables (compact variables) is designated below by the letter
Ω, i.e. Ω = (η2 , η3 , . . . , ηNe , ω1 , ω2, . . . , ωNe ). Analogously, the partial set of angular varibles
is designated below by the letters Ωj (= η2 , η3 , . . . , ηj , ω1 , ω2 , . . . , ωj ) for j = 2, 3, . . . , Ne and
ΩNe = Ω. These angular variables describe all angular configurations in the cluster of j
bound electrons. The square of the generalized angular momentum operator for the cluster
of j bound particles/electrons is defined by the following recursive relation
∂2 (3j − 4) cos2 ηj − 2 sin2 ηj ∂ Λ2j−1(Ωj−1 ) ~ℓ2
j
Λ2j (Ωj ) =− 2 − + 2 + 2
(15)
∂ηj sin ηj cos ηj ∂ηj sin ηj cos ηj
with the following ‘initial’ condition: Λ21 (Ω1 ) = ~ℓ21 (ω1 ).
The 3Ne dimensional ‘total’ Laplacian has a very simple form in the hyperspherical
coordinates
Ne
∂2 3Ne − 1 ∂ Λ2Ne (Ω)
∇2Ne = ∇2i=1 =
X
+ − (16)
i=1 ∂r 2 r ∂r r2

9
This term is proportional to the kinetic energy of an atom/ion which contains Ne bound
electrons (see below). The definition of the hyperspherical coordinates is completed by spec-
ifying the volume element in this coordinates dτ = r 3Ne −1 drdΩ, where dΩ is the differential
surface area on the 3Ne −dimensional hypersphere, i.e.
Ne Ne Ne Ne
(cos2 ηj sin3j−4 ηj dηj ) (cos2 ηj sin3j−4 ηj dηj )
Y Y Y Y
dΩ = (sin θi dωi ) = (sin θi dθi dφi ) (17)
j=2 i=1 j=2 i=1

More detail description of the hyperspherical coordinates and analysis of their properties
can be found, e.g., in [13] and in a large number of papers, books and textbooks on the
method of hyperspherical harmonics and its applications to different problems known in
atomic, molecular and nuclear physics (see, e.g., [14] - [19] and references therein).
The (2Ne −1) Laplace operators Λ2Ne (Ω), . . . , Λ2j (Ωj ), . . . , Λ22(Ω2 ), ~ℓ21 (ω1 ), ~ℓ22 (ω2 ), . . . , ~ℓ2Ne (ωNe )
depend upon different sets of angular variables. Therefore, these operators commute with
each other and they have a common system of eigenfunctions. These eigenfunctions are rep-
resented in the form of products of eigenfunctions of these (2Ne −1) Laplace operators. These
eigenfunctions can be chosen as the ‘natural’ basis set in the (3Ne − 1) angular (compact)
space Ω. It is clear that each of these basis functions includes the product of the spherical
harmonics of each electron, i.e. Y(Ω) ∼ Yℓ1 m1 (θ1 , φ1 )Yℓ2 m2 (θ2 , φ2 ) . . . YℓNe mNe (θNe , φNe ).
The eigenfunctions of the Ne − 1 hyperspherical angles η2 , η3 , . . . , ηNe are the polynomial
functions which are usually expressed in terms of the Jacobi (spherical) polynomials
Pn(α,β) (x) [20], [21]. The products of eigenfunctions of all (2Ne − 1) differential operators
Λ2 (Ω), . . . , Λ2 (Ωj ), . . . , Λ2 (Ω2 ), ~ℓ2 (ω1 ), . . . , ~ℓ2 (ωNe ) mentioned above which depend upon
Ne j 2 1 Ne

2Ne angular and Ne − 1 hyperangular variables are called the hyperspherical harmonics,
or HH functions. In this study to designate the HH functions we use the notation
YK(b),
~ ~
~
(Ω), where K(b), ~ℓ(b), m(b)
~ is the multi-index of the hyperspherical harmonics.
ℓ(b),m(b)
~

The numerical value of each component of this multi-index is uniformly related with the
eigenvalues of the corresponding Laplace operators mentioned above.
In actual atomic computations only those hyperspherical harmonics (HH) are important
which have the correct permutations symmetry between all bound electrons. In some earlier
works these hyperspherical harmonics were called the ‘physical’ (or actual) HH. For atomic
systems the physical harmonics can be constructed, e.g., with the use of the projection
h i
SSz
operators PLM π for the given atomic term L, M, S, Sz , π . The explicit construction of such

projectors is well described in a number of original papers. For simple atomic systems,

10
e.g., for the two-electron atoms/ions the explicit construction of such projection operators is
simple (see, e.g., [22]). The physical hyperspherical harmonics are extensively used in various
problems of few-body physics, including description of many different atomic systems (see,
e.g., [18] and references therein).

IV. FACTORIZATION METHOD FOR FEW- AND MANY-ELECTRON ATOMS

AND IONS

In hyperspherical coordinates the Hamiltonian of an atom which contains Ne bound

electrons is written in the form [11], [12] (see, also [13])

1 h ∂2 3Ne − 1 ∂ Λ2Ne (Ω) i W (Ω)

H(r, Ω) = − + − + (18)
2 ∂r 2 r ∂r r2 r

where Λ2Ne (Ω) is the hypermomentum of the atom, while W (Ω) is the hyperangular part of
the interaction (Coulomb) potential which includes electron-nucleus and electron-electron
parts. For an atom with Ne bound electrons the electron-nucleus term contains Ne terms,
Ne (Ne −1)
while the electron-electron part includes the 2
terms. Now, we can consider the
matrix of the operator H(r, Ω) in the basis of hyperspherical harmonics (or HH-basis, for
short), i.e.

Ĥab (r) = hYK(a),

~ ~ ℓ(a),m(a)
~
(Ω) | H(r, Ω) | YK(b),
~ ~
ℓ(b),m(b)
~ (Ω)i (19)

where YK(c),
~ ~ (Ω) are the physical hyperspherical harmonics (see above), K(c) ~ =
ℓ(c),m(c)
~

(K1 , K2 , . . . .Kg ), ~ℓ(c) = (ℓ1 , ℓ2 , . . . , ℓg ) and m(c)

K ℓ
~ = (m1 , m2 , . . . , mgm ) are the multi-indeces
(or vector-indeces) which uniformly define the hyperspherical harmonics YK(a),
~ ~ ℓ(a),m(a)
~
(Ω).
In turn, these multi-indeces of the hyperspherical harmonics are determined by the atomic
state (or atomic term) considered in calculations. In actual computations the dimensions
of these vector-indeces gK , gℓ and gm should be minimal, since all hyperspherical harmonics
applied in numerical computations are the physical HH. This means that these HH have
the correct permutation symmetry, or, in other words, correct symmetry in respect to the
required permutations of all electron indeces. For instance, the hyperspherical harmonics
which are needed in bound state calculations of the singlet 1 S(L = 0)−states of the he-
lium atom are written in the form | K, ℓ, ℓi =| K, ℓi, where K = 0, 2, 4, . . . , 2n is their
K
hypermomentum (index) , while ℓ ≥ 0 (and ℓ ≤ 2
) is athe second index (more details can

11
be found in [22]). In other words, for this atomic system each physical HH is designated
by the two-component multi-index (K, ℓ), i.e. in the notations introduced above one finds
gK = 1, gℓ = 1 and mgm = 0. Below, we shall designate the hyperspherical matrix of the
Hamiltonain Ĥab (r) by using the same notation H, or H(r) (as we did in the second Section).
It should be mentioned that H(r) is the differential operator in respect to the hyper-radius
r of the second order. The explicit form of the H(r) Hamiltonian operator is

1 h ∂2 3Ne − 1 ∂ K̂(K̂ + 3Ne − 2) i Ŵ

H(r, Ω) = − + − + (20)
2 ∂r 2 r ∂r r2 r
In our earlier study [17] we have shown that the matrix of the atomic Hamiltonian H(r),
Eq.(20), which contains Ne bound electrons is always factorized, i.e. it is represented in the
form

H = Θ∗1 (r)Θ1 (r) + â1 (21)

where â1 is a matrix defined below, while the operator Θ1 (r) and its adjoint operator Θ∗1 (r)
are the first-order differential operators defined as follows

1 h ∂ 3Ne − 1  β̂1 i
Θ1 (r) = √ − + + + α̂1 (22)
2 ∂r 2r r
and
1 h ∂ 3Ne − 1  β̂1 i
Θ∗1 (r) = √ + + + α̂1 (23)
2 ∂r 2r r
where the notations β̂1 , α̂1 and â1 in Eqs.(21) - (23)) stand for the symmetric, infinite-
dimensional, in principle, matrices which do not commute with each other. In actual appli-
cations the dimensions of these matrices coincide with the total number of hyperspherical
harmonics used. By substituting these two expressions, Eqs.(22) - (23), into Eq.(21) one
finds the three following equations for the α̂1 , β̂1 and â1 matrices:
 3Ne − 1  3Ne − 1 
β̂1 (β̂1 − 1) = K̂ + K̂ + −1 (24)
2 2
α̂1 β̂1 + β̂1 α̂1 = 2Ŵ (25)
1
â1 = − α̂12 (26)
2
where the matrix of hypermomentum K̂ is a diagonal matrix in the basis of hyperspherical
harmonics (or, in K−representation, for short). Solution of Eq.(24) is written in the form
3Ne − 1
β̂1 = K̂ + (27)
2

12
where we use the fact that the atomic wave function is regular at r = 0, or at the atomic
nucleus. As follows from this equation the matrix β̂1 is diagonal in K−representation.
Below, we apply only this K−representation, since it substantially simplifies a large number
of formulas derived below. In particular, by using Eq.(25) and the formula from [23] (see
Chapter 10, $ 18) we can write the explicit expression for the α̂1 matrix
Z +∞
α̂1 = 2 exp(−β̂1 t)Ŵ exp(−β̂1 t)dt (28)
0

Since the β̂1 matrix is diagonal, then for the (ij)−matrix element of the α̂1 matrix takes the
form
h i 2Wij 2Wij 2Wij
α̂1 = = = (29)
ij [β1 ]ii + [β1 ]jj [β1 ]i + [β1 ]j Ki + Kj + 3Ne − 1
Finally, we can determine the â1 matrix from Eq.(26). In particular, for the (ij)−matrix
elements of the â1 matrix one finds
h i X Wik Wkj X 1 h i 1
â1 = −2 · = −2 Wik Wkj (30)
ij
k βi + βk βk + βj k βi + βk βk + βj

At the second stage of the procedure, we introduce the set of radial operators Θn (r),
where n = 2, 3, . . ., which are similar to the operator Θ1 (r) defined above (see, Eq.(22)), i.e.

1 h ∂ 3Ne − 1  β̂n i
Θn (r) = √ − + + + α̂n (31)
2 ∂r 2r r

Therefore, its adjoint operator takes the form

1 h ∂ 3Ne − 1  β̂n i
Θ∗n (r) = √ + + + α̂n (32)
2 ∂r 2r r

In order to construct the correct and logically closed algorithm of the factorization method
the following conditions must be obeyed

Θn (r)Θ∗n (r) + ân = Hn+1 = Θ∗n+1 (r)Θn+1 (r) + ân+1 (33)

for n = 1, 2, . . .. By substituing the explicit expressions, Eqs.(31) and (32) into Eq.(33) we
obtain the following equations for the β̂n , β̂n+1 , α̂n , α̂n+1 , ân and ân+1 matrices

β̂n+1 (β̂n+1 − 1) = β̂n (β̂n + 1) , (34)

α̂n βn + βn α̂n = 2Ŵ = αn+1 βn+1 + βn+1 α̂n+1 , (35)

1 1 2
ân = − αn2 , ân+1 = − αn+1 (36)
2 2

13
These matrix equations look very similar to the analogus numerical equations mentioned
in Section II (see, Eqs.(9)). However, these equations Eqs.(34) - (36), are written for the
symmetric, infinte-dimensional matrices, which do not commute with each other, e.g., the
β̂n matrix do not commute with the α̂n and ân+1 matrices, etc. Solution of these equations,
Eqs.(34) - (36), regular at r = 0 is written in the following form(s)
3Ne − 1
β̂n+1 = β̂n + 1 = . . . = β̂1 + n = K̂ + +n (37)
2
Z +∞
α̂n+1 = 2 exp(−β̂n+1 t)Ŵ exp(−β̂n+1 t)dt (38)
0
1 2
ân+1 = − αn+1 (39)
2
Note that the second equaition here (Eq.(38)) produces the following explicit expression
for the (ij)−matrix element of the α̂n+1 matrix
h i 2Wij 2Wij 2Wij
α̂n+1 = = = (40)
ij [βn+1 ]ii + [βn+1 ]jj [β1 ]i + [β1 ]j + 2n Ki + Kj + 3Ne − 1 + 2n

where [β1 ]i is the (ii)−matrix element of the diagonal β̂1 matrix, i.e. [βn+1 ]ij = δij [βn+1 ]ii =
δij [βn+1 ]i and [β1 ]ij = δij [β1 ]ii = δij [β1 ]i . This leads to the following analytical formula for
the (ij)−matrix elements of the ân+1 matrix
h i X Wik Wkj
ân+1 = −2 ·
k [β1 ]i + [β1 ]k + 2n [β1 ]k + [β1 ]j + 2n
ij

1X 1 h i 1
= − Wik Wkj (41)
2 k bik + n bkj + n
X 1 h i 1
= −2 Wik Wkj
k Ki + Kk + 2n + 3Ne − 1 Kk + Kj + 2n + 3Ne − 1

where bik = 12 ([β1 ]i + [β1 ]k ) and bkj = 21 ([β1 ]k + [β1 ]j ), while Ki are the matrix elements
of the diagonal K̂-matrix (the matrix of hypermomentum) and n ≥ 0, where n is the
radial quantum number (integer, non-negative). Formally, the formula, Eq.(41), is a direct
generalization of the Bohr’s formula, originally derived by N. Bohr (in 1913) for the hydrogen
atom, to an atom/ion which contains Ne bound electrons. In Quantum Mechanics the same
formula for the spectra of the hydrogen atom was derived by W. Pauli in 1926 [24]. For
Ne = 1 the formula Eq.(41) exactly coincides with the formula Eq.(3) (in atomic units).
As follows from Eq.(39) for any given bound state in many-electron atoms the radial
quantum number n is a conserving quantum number which can be used to number (or
locate) this bound state inside of one series of bound states which have the same values

14
of L, Lz (orM), S, Sz and π. In other words, the radial quantum number n (in our nota-
tion) can be used to number the bound states inside of one atomic term. In general, any
bound state in the atomic term can be designated with the use of the following notation
h i h i
| n, L, M, S, Sz , π i, where the internal notation L, M, S, Sz , π designates the correspond-
ing atomic term and n is the number of this (bound) state in this atomic term, or, in
other words, the number of excitation(s). The same notation can be used to designate the
corresponding wave function(s).

V. BOUND STATE WAVE FUNCTIONS

Let us discuss an approach which can be used to determine the wave functions of the
bound states in atoms/ions which contain Ne bound electrons. This approach is based on
the basic equations of matrix mechanics derived above and has a number of similarities with
the analogous method used in Section II for one-electron atoms and ions. In particular, the
ground (bound) state wave functions can be determined from the differential equation of
the first order Θ1 (r)Ψ(r) = 0. The explicit form of this equation is
h  ∂ 3Ne − 1  β̂1 i h ∂ K̂ i
− + + + α̂1 Ψ(r) = − + + α̂1 Ψ(r) = 0 (42)
∂r 2r r ∂r r
where K̂ is the matrix of the hypermomentum, i.e. K̂ = β̂1 − 3N2e +1 . From Eq.(42) we obtain
the following equation
h ∂ K̂ i
− + + α̂1 Ψ(r) = 0 (43)
∂r r
To solve this equation we can represent the function Ψ(r) in the form Ψ(r) = r K̂ exp(λr)C,
where λ is a real (always negative) numerical constant defined below and C is the numer-
ical vector, i.e. each component of this vector does not depend upon the hyper-radius r.
Subsistution of the function Ψ(r) in this form into Eq.(43) reduces this equation to the form
h K̂ K̂ i h i
− −λ+ + α̂1 Ψ(r) = α̂1 − λ r K̂ exp(λr)C = 0 (44)
r r
In other words, to determine the numerical value of λ we need to solve the following gen-
eralized eigenvalue problem: ( − λB̂)C = 0, where the matrix elements of the  and B̂
matrices are defined by the following equations
1 +∞
Z  
[Â]ij = r K i α1 r Kj exp(2λr)r 3Ne −1 dr (45)
Ni Nj 0 ij

15
 δij
Z +∞
[B̂]ij = 2 r 2Ki exp(2λr)r 3Ne−1 dr (46)
Ni 0

As follows from the second equations the matrix B̂ is diagonal and all its eigenvalues (i.e.
diagonal elements) are positive. In general, we can choose the normalization constants Ni
in such a manner that the matrix B̂ will coincide with the unit matrix. This reduces the
ˆ − λ)C = 0, where Ã
original problem to the form of a regular eigenvalue problem, i.e. (Ã ˆ is

the matrix  in the new ‘normalized’ basis.

This eigenvalue problem is equivalent to the finding of the absolute minimum of the
following energy functional
hΨ | H | Ψi hΨ | Θ∗1 (r)Θ1 (r) | Ψi hΨ | â1 | Ψi
E1 (Ψ) = = +
hΨ | Ψi hΨ | Ψi hΨ | Ψi
hΘ1 (r)Ψ | Θ1 (r)Ψi hΨ | â1 | Ψi
= + (47)
hΨ | Ψi hΨ | Ψi
In our strategy of minimization the trial wave functions is represented in the form of the
PN
linear combinations: Ψ(r) = r K̂ exp(λr)C = n=1 Cn r K̂n exp(λr) = Cr K̂n exp(λr). The
first term in the right-hand side of Eq.(47) is always positive. For our choice of the radial
trial functions this term equals zero identically. The second term in the right-hand side of
Eq.(47) is always negative. Optimization of the linear coefficients Ck in our wave function
at the second stage of the method means that we are trying to make the second term in
Eq.(47) as negative as possible.
Note that the matrix elements of the both  and B̂ matrices defined by Eqs.(45) - (46)
depend upon the eigenvalue λ which is not known at the first stage of the solution. Formally,
we can start the process by using some λ0 value which is close to the exact answer. For atoms
and ions the numerical value of λ0 can be taken from accurate variational computations of
the atoms/ions performed with the use of the multi-configuration Hatree-Fock mehtod [25],
or by applying highly accurate methods for calculations of the total energies of the few-
electron atoms and ions. The chosen numerical value of λ0 is used in Eqs.(45) - (46) to
calculate the both Â(λ0 ) and B̂(λ0 ) matrices. At the next step we solve the generalized
eigenvalue problem (Â − λB̂)C = 0 and determine the new ‘improved’ eigenvalue λ1 which
is closer to the actual (or exact) eigenvalue λ, than λ0 . Then with the use of this λ1 value
we compute the new Â(λ1 ) and B̂(λ1 ) matrices an solve the (Â − λB̂)C = 0 equation. This
leads to the new eigenvalue λ2 which is better that the λ1 value. Then the process can be
repeated as many times as needed to reach a required numerical accuracy for the actual

16
eigenvalue λ. The wave functions of the first and other excited state in the atoms/ions with
the Ne bound electrons are determined analogously, and we do not want to discuss these
cases here.

VI. CONCLUSION

We have developed the method of matrix factorization which can be applied to many-
electron (or many-particle) atoms, ions and molecules. Formally, this method can be used
for arbitrary many-body systems where each pair of particles interacts by the Coulomb
potential. Briefly, for each of these systems the corresponding Hamiltonian written in the
hyperspherical multi-dimensional coordinates must be similar to the form represented by
Eq.(18) ([11] - [13]). The main difference between the matrix factorization and ‘regular’ (or
numerical) factorization follows from the fact that in the method of matrix factorization
we use a number of infinite-dimensional matrices which do not commute with each other.
This fact complicates the procedure of matrix factorization and its applications to many-
electron atomic systems. Nevertheless, we could develop the closed algorithm of the matrix
factorization, and now this method can be applied to determine the bound states in a large
number of actual (i.e. few- and many-electron) atomic systems.
The method of matrix factorization allows one to determine the bound state spectra of
many-electron (but non-relativistic!) atoms, ions and molecules. This means that by using
this method one can determine, in principle, all bound state energies and corresponding wave
functions. At the following stages these wave functions can be appled to evaluate various
bound state properties, including lowest-order relativistic and QED corrections for different
atoms, ions and molecules. Formally, the method of matrix factorization allows one to obtain
analytical and semi-analytical answers to numerous questions about atomic structure of the
few- and many-electron (non-relativistic) atoms, ions and light molecules. In many cases,
however, the obtained answers and solutions are often written in the matrix form which is
directly related to the original matrix form of the matrix (quantum) mechanics.
Unfortunately, this paper has not ben published in the middle of 1950’s, or even earlier,
when Dirac, Fock and Heisenberg were around. Finally, our method of matrix factorization
of the Coulomb many-particle Hamiltonians has been developed with a substantiall time
delay. At the same time a large number of competing computational methods have exten-

17
sively been developed and applied in atomic physics. Some of these methods became very
effective, relatively simple and fast procedures. However, the method of matrix factoriza-
tion has a great potential for future development and various modifications, since it is based
on the internal ‘ladder’ structure of the Coulomb Hamiltonians. Furthermore, the matrix
factorization is the new, relatively simple and advanced approach which can be used to
investigate the few- and many-body Coulomb problems and determine the bound states in
such systems. In particular, our method can be used to understand some interesting details
of atomic spectra and substantially simplify accurate bound state computations of different
systems known in atomic and molecular physics.
Finally, we want to empasize that the method of matrix factorization is substantially
based on the ladder stucture of the Hamiltonians of the Coulomb many-body systems.
In this study we discovered the method which uses this ladder structure of the Coulomb
Hamiltonians and allows one to determine all bound states in any few- and/or many-body
Coulomb system. Based on the ladder stucture of the Coulomb Hamiltonians we can predict
that this method will be a very effective tool for theoretical and numerical investigation of
such spectra. For instance, the method of matrix factorization allows one to study general
dependencies of the total energies of different bound states in the few- and many-electron
atoms/ions upon good quantum numbers a priory known for such quantum systems.

[1] H.S. Green (1965), Matrix Mechanics [P. Noordhoff Ltd, Groningen, Netherlands (1965)],
ASIN: B0006BMIP8.
[2] W. Heisenberg, Zeits. für Physik, 33, 879 (1925).
[3] M. Born, W. Heisenberg and P. Jordan, Zeits. für Physik, 35, 557 (1925).
[4] P.A.M. Dirac, The Principles of Quantum Mechanics (4th ed., Oxford at the Clarendon Press,
Oxford (UK) (1958)).
[5] L.D. Landau and E.M. Lifshitz, Quantum Mechanics: Non-Relativistic Theory, (3rd. ed. Perg-
amon Press, New York (1989)).
[6] P.V. Elutin and V.D. Krivchenkov, Quantum Mechanics with Problems, (Nauka (Science),
Moscow (1976)), Chpts. III and V (in Russian).
[7] S.T. Epstein, The Variation Method in Quantum Chemistry, (Academic Press, New York

18
 (1974)).
[8] I.M. Gel’fand, Lectures om Linear Algebra, (Dover Publications, Inc., New York (1993)), Chpt.
II.
[9] P.R. Halmos, Finite-Dimeesional Vector Spaces, (Springer-Verlag, New-York (1987)).
[10] I.I. Sobelman, Introduction to the Theory of Atomic Spectra, (Nauka (Science), Moscow
(1972)).
[11] V.A. Fock, Izv. Akad. Nauk SSSR, Ser. Fiz. 18, 161 (1954).
[12] Yu.N. Demkov and A.M. Ermolaev, Sov.-Phys. JETP, 9, 633 (1959).
[13] Dw.L. Knirk, J. Chem. Phys. 60, 66 (1974).
[14] B.A. Fomin and V.D. Efros, Sov.-Phys. Nucl. Phys., 34, 455 (1981).
[15] P.C. Abbot and E.N. Maslen, J. Phys. B, 17, L489 (1984).
[16] R. Schneider, Phys. Lett. B, 40, L439 (1972).
[17] A.M. Frolov, J. Phys. B 19, 2041 (1986).
[18] J.S. Avery, J. Comput. Phys. and Applied Mathematics, 233, 1366 (2010).
[19] E. Pelikan and H. Klar, Zeits. für Physik A, 310, 153 (1983).
[20] I.S. Gradstein and I.M. Ryzhik, Tables of Integrals, Series and Products, (6th revised ed.,
Academic Press, New York (2000)).
[21] Handbook of Mathematical Functions (M. Abramowitz and I.A. Stegun (Eds.), Dover, New
York, 1972).
[22] V.D. Efros, A.M. Frolov and M.I. Mukhtarova, J. Phys. B 15, L819 (1982).
[23] R. Belman, Introduction to Matrix Analysis, (McGraw-Hill Book Company, Inc., New York
(1960)).
[24] W. Pauli, Zeits. für Physik, 36, 336 (1926).
[25] C. Froese Fisher, T. Brage and P. Jonsson, Computational Atomic Structure. An MCHF
Approach (IOP Publishing, Bristol UK, 1997).

19